Reference image selection for difference imaging analysis

Difference image analysis (DIA) is an effective technique for obtaining photometry in crowded fields, relative to a chosen reference image. As yet, however, optimal reference image selection is an unsolved problem. We examine how this selection depends on the combination of seeing, background and detector pixel size. Our tests use a combination of simulated data and quality indicators from DIA of well-sampled optical data and under-sampled near-infrared data from the Optical Gravitational Lensing Experiment and VVV surveys, respectively. We search for a figure-of-merit (FoM) which could be used to select reference images for each survey. While we do not find a universally applicable FoM, survey-specific measures indicate that the effect of spatial under-sampling may require a change in strategy from the standard DIA approach, even though seeing remains the primary criterion. We find that background is not an important criterion for reference selection, at least for the dynamic range in the images we test. For our analysis of VVV data in particular, we find that spatial under-sampling is best handled by reversing the standard DIA procedure and convolving target images to a better-sampled (poor-seeing) reference image.